In order to complete the processing using a computer in as a short time as possible, it is indispensable to utilize a super computer having high processing performance. However, since the super computer having the high processing performance is very expensive, a computer having lower processing performance is often used to make processing. As a result, it takes considerable time to make processing and when the processing capability thereof is exceeded, there is the possibility that the computer hangs up on the way of processing.
Recently, attention is paid to the technique named grid computing as a measure for solving such problem. “Grid” is a word derived from the “power grid” (high-voltage transmission network). The grid computing is the environment in which computers dispersed geographically are combined to make it possible to utilize computer resources without consciousness of occurrence places thereof as if they were water or electricity and forms a different-kind mixing system constituted by a plurality of computers. Further, the grid computing is the setup in which execution environment of data and programs can be shared mutually among systems.
The grid computing makes it possible to utilize resources of high-performance computers in the grid so that processing of a heavy load can be also executed even when there is not provided high-performance computers in itself. Further, the grid computing can divide processing into units that can be independently executed in parallel and assign the units to a plurality of computers existing in the grid, respectively, to thereby disperse a load. Accordingly, even if computers having relatively low performance are combined, processing of a heavy load can be executed.
Moreover, in the system having a plurality of computers having high performance, a load can be dispersed properly by the grid. In other words, when a load is not dispersed consciously, there is a tendency that a load is concentrated on a computer having highest performance and other relatively low-speed computers fall into the situation that they do not almost use the CPU power. Accordingly, each of the computers provided in the system can be managed as a member of the grid to thereby disperse the load and utilize resources efficiently.
From the above advantages, the computer grid is constructed as the computer environment realized by integrating techniques such as virtual operating system (OS), clustering, distributed system and autonomous system to produce results. The setup for constructing the grid computing is mainly formed in a middleware level so far.
The advantages of the computer grid are considered to be obtained not only in the scientific and technical fields but also in the business grid in which the computer grid is applied to the business field and which is represented in that the business application service, for example, is provided efficiently.
In a data center divided into an enterprise system for operating the business application to process requests from users and a management system for managing servers and the like constituting the enterprise system and in which the enterprise system and the management system are connected through a network to each other, addition or exchange of enterprise servers constituting the enterprise system due to deterioration of the enterprise servers and increased customers is indispensable and important operation and management business. In such operation and management business, it is important to grasp the performance of newly introduced enterprise servers when environment of the enterprise system is set and an amount of resources such as the number of enterprise servers is decided.
As an apparatus for measuring performance of a plurality of servers efficiently, a server performance measurement apparatus is disclosed in patent document 1 (JP-A-2002-41331). The server performance measurement apparatus includes a measurement condition setting unit for setting performance measurement conditions, a performance measurement execution control unit for broadcasting a performance measurement instruction produced on the basis of the performance measurement conditions to all servers to be performance-measured to thereby measure performance of the servers on the same conditions and a measured result management unit for managing measured results transmitted from the servers collectively.
Further, as an apparatus for measuring performance and load of a communication apparatus in actual operation efficiently, a performance and load measurement apparatus is disclosed in patent document 2 (JP-A-2004-171226). The performance and load measurement apparatus includes a measurement unit for measuring the reception time that a plurality of packets processed by the communication apparatus successively are received and a performance presumption unit for presuming packet processing performance of the communication apparatus on the basis of the time intervals of the reception time of the packets measured by the measurement unit, to thereby measure the performance or load of the communication apparatus.
As an apparatus for providing a method for reproducing an access pattern near to actual operation environment of WWW server to obtain marginal performance value having high reliability, a marginal performance measurement apparatus of WWW server is disclosed in patent document 3 (JP-A-2004-318454). The marginal performance measurement apparatus of WWW server, after reproducing a load based on a recorded access pattern of an end user for WWW server in actual operation environment, reproduces the load while increasing the access frequency to measure performance value, so that the access frequency when predetermined requirement performance of WWW server is exceeded is presented as the marginal performance value of WWW server.
Further, the benchmark test is also known as a method for estimating server performance. There is known, for example, the performance measurement tool SPECjAppServer 2004 (Trade Name) of EJB (Enterprise Java Beans) server by Web 3-hierarchy application in which EJB application is connected to processing such as JSP (Java Server Pages), Servlet and JDBC driver and the load test is performed to calculate the workload of the EJB server.
However, in the data center in which the business application is operated to process the requests from the users, even when the server performance is measured, the estimation standard for server performance is different depending on characteristics of business applications to be operated and accordingly it is necessary to measure the server performance in each business application to be operated.
The server performance measurement apparatus disclosed in the patent document 1 can measure performance of a plurality of servers at the same time zone, although since a concrete performance measurement method is not described, it is considered that performance cannot be measured for each business application.
The performance and load measurement apparatus disclosed in the patent document 2 is to measure performance and load of the communication apparatus in actual operation efficiently without requiring human labor due to installation and management of apparatuses, although since performance and load of servers are presumed from the transmission and reception time of packets constituting test data, it is impossible to measure performance for each business application.
The marginal performance measurement apparatus of WWW server disclosed in the patent document 3 is to utilize the access pattern of the end user near to the actual operation environment of WWW server to measure performance of WWW server, although since the access pattern of the end user stored in database is not formed on condition that every state is supposed and the measured performance is the marginal performance value, it is difficult to get a performance index reflecting characteristics of each business application.
Therefore, there is a problem that the conventional method and apparatus for measuring the server performance cannot obtain the performance index reflecting characteristics of each business application.
Further, in the estimation method of server performance using the benchmark test, a load is given to the server to be measured externally until the load on the server reaches 100% so that performance thereof is measured and accordingly there is a problem that the server performance index reflecting characteristics of each business application cannot be obtained.